Energy Transfer from High-Shear, Low

advertisement
1524
JOURNAL OF PHYSICAL OCEANOGRAPHY
VOLUME 42
Energy Transfer from High-Shear, Low-Frequency Internal Waves
to High-Frequency Waves near Kaena Ridge, Hawaii
OLIVER M. SUN AND ROBERT PINKEL
Marine Physical Laboratory, Scripps Institution of Oceanography, La Jolla, California
(Manuscript received 17 June 2011, in final form 23 March 2012)
ABSTRACT
Evidence is presented for the transfer of energy from low-frequency inertial–diurnal internal waves to highfrequency waves in the band between 6 cpd and the buoyancy frequency. This transfer links the most energetic
waves in the spectrum, those receiving energy directly from the winds, barotropic tides, and parametric
subharmonic instability, with those most directly involved in the breaking process. Transfer estimates are
based on month-long records of ocean velocity and temperature obtained continuously over 80–800 m from
the research platform (R/P) Floating Instrument Platform (FLIP) in the Hawaii Ocean Mixing Experiment
(HOME) Nearfield (2002) and Farfield (2001) experiments, in Hawaiian waters. Triple correlations between
low-frequency vertical shears and high-frequency Reynolds stresses, huiw›Ui/›zi, are used to estimate energy
transfers. These are supported by bispectral analysis, which show significant energy transfers to pairs of waves
with nearly identical frequency. Wavenumber bispectra indicate that the vertical scales of the high-frequency
waves are unequal, with one wave of comparable scale to that of the low-frequency parent and the other of
much longer scale. The scales of the high-frequency waves contrast with the classical pictures of induced
diffusion and elastic scattering interactions and violates the scale-separation assumption of eikonal models
of interaction. The possibility that the observed waves are Doppler shifted from intrinsic frequencies near f
or N is explored. Peak transfer rates in the Nearfield, an energetic tidal conversion site, are on the order of
2 3 1027 W kg21 and are of similar magnitude to estimates of turbulent dissipation that were made near the
ridge during HOME. Transfer rates in the Farfield are found to be about half the Nearfield values.
1. Introduction
Internal wave energy and shear in the oceans are
concentrated near the low-frequency, inertial end of
the spectrum. Evidence for the direct breaking of these
energetic low-frequency waves is limited. Klymak et al.
(2008) have observed the convective instability of the
semidiurnal baroclinic tide immediately above its generation site at Kaena Ridge, Hawaii. Alford and Gregg
(2001) describe the dissipation of a low-latitude inertial
wave by Kelvin–Helmholtz instability. In general, it is
thought that energy is first transferred from low to
higher-frequency internal waves and that these waves
subsequently break (e.g., Alford and Pinkel 2000a,b).
Resonant interaction theory, beginning with Phillips
(1960) and developed by Hasselmann (1962), Benney
and Saffman (1966), McComas and Bretherton (1977),
McComas and Müller (1981), Müller et al. (1986), and
Corresponding author address: Oliver M. Sun, Woods Hole
Oceanographic Institution, M.S. 21, Woods Hole, MA 02543.
E-mail: osun@whoi.edu
DOI: 10.1175/JPO-D-11-0117.1
Ó 2012 American Meteorological Society
many others, predicts energy transfers between triads
of internal waves with wave vectors (k1, k2, k3) and
frequencies (v1, v2, v3), which together satisfy the resonance conditions
k1 1 k2 1 k3 5 0 and
v1 1 v2 1 v3 5 0:
(1)
(2)
Each wave also approximately satisfies the linear dispersion relationship
v2 5
f 2 m2 1 N 2 (k2 1 l2 )
,
k2 1 l2 1 m2
k 5 (k, l, m),
(3)
where N is the Brunt–Väisälä frequency and f is the
inertial frequency.
Evaluations of energy transfer rates within a Garrett–
Munk (GM) spectrum (Munk 1981) found that scaleseparated interactions are likely to be important because
of the wavenumber slope of the spectrum. Three limiting cases of scale separation were identified by McComas
SEPTEMBER 2012
SUN AND PINKEL
and Bretherton (1977), who named the interactions
induced diffusion (ID), elastic scattering (ES), and
parametric subharmonic instability (PSI). ID involves
an energy exchange between two high-frequency waves
that have small spatial scales (large wavenumbers) and
a third, much lower–frequency wave with large spatial
scale (small wavenumber). Because the large wave vectors of the high-frequency waves differ by the small wave
vector of the low-frequency wave, ID shifts wave action
by a small distance in wavenumber space. In the limit of
large-spatial-scale separation, wave action tends to be
diffused in wavenumber space. ES also connects a lowfrequency wave with a pair of high-frequency waves.
The high-frequency waves have nearly opposite vertical
wavenumber, with about half the wavelength of the
low-frequency wave, and hence the process is reminiscent of Bragg scattering. The ES triad is not widely
separated in vertical wavenumber, but the horizontal
wavenumbers (and hence the frequencies) of the scattered waves may be large. In PSI, unlike the other two
triads, energy is transferred between one wave and a pair
of nearly half-frequency waves. These subharmonics
have nearly opposite wave vectors (in both the horizontal and the vertical) and thus may have large vertical wavenumbers relative to the original.
Resonant interaction theory is exact only in the weakly
nonlinear limit. Eikonal or ray-tracing methods have
been also used to model interactions of small-scale test
waves as they pass through and are refracted by a background field of large-scale waves (Henyey and Pomphrey
1983; Henyey et al. 1986; Broutman and Young 1986).
This model allows stronger background fields than in
the resonant theory, but it assumes, rather than demonstrates, a scale separation between the test waves and
the background. [Henyey et al. (1986) find that the
results remain qualitatively valid up to ‘‘scale equivalence.’’] Also, the test waves interact only with the
background and not one another, so that the notion of
a resonant triad is lost.
Measured open-ocean turbulence dissipation rates
are reasonably consistent with the predictions of resonant interaction theory (Henyey et al. 1986; Gregg
1989; Polzin et al. 1995). However, direct observations
of nonlinear energy transfer to high-frequency internal
waves (e.g., statistically significant bicoherences showing phase coupling between triads of waves) are few. The
slowness of the energy transfers in a ‘‘typical,’’ near-GM
ocean has been viewed as a significant obstacle to
achieving statistically significant observations over
the length of a feasible field experiment (McComas
and Briscoe 1980). A few studies have examined correlations between high-frequency wave stresses and
low-frequency shears that are likely associated with
1525
near-inertial waves (Jacobs and Cox 1987; Duda and
Jacobs 1998).
The present study documents nonlinear interactions between low- and high-frequency internal waves
near Kaena Ridge, Hawaii, a site of intense barotropicbaroclinic conversion. We use data collected aboard
the research platform (R/P) Floating Instrument Platform
(FLIP) during 2001–02 as part of the Hawaii Ocean Mixing Experiment (HOME) (Rudnick et al. 2003; Rainville
and Pinkel 2006a,b). The FLIP data combine CTD and
Doppler sonar measurements to provide profiles of temperature, density, and horizontal and vertical velocity,
covering a depth range of 80–800 m with 2–4-m vertical
resolution. The 4-min profiling interval allows us to resolve the entire frequency range for internal waves, from
the inertial frequency to the buoyancy frequency. We
attempt to quantify nonlinear energy transfer rates
through the triple correlation between high-frequency
wave stresses and low-frequency wave shears. Bispectral
analysis is used to identify specific frequencies and wavenumbers of triads whose interactions contribute to the
energy transfers.
In the next section, we begin with a brief description
of the instrumentation. Samples of the depth–time series
of isopycnal displacements and horizontal velocities are
presented and contrasted with vertical velocities and
vertical shears. Spectral estimates are presented to establish the important frequencies and scales of motion.
We then develop the basis for the triple correlation
method and use it to estimate energy transfers at both
HOME locations. Bispectral methods are explained and
used to decompose the triple correlations by frequency
and wavenumber. We attempt to sketch a picture of the
interacting waves. Finally, we attempt to summarize the
results and compare them to existing estimates of energy transfer and dissipation. A major result of the
analysis is that a significant energy transfer from lowfrequency to high-frequency waves is detected in both
HOME Nearfield and Farfield. The interactions do not
fit with preexisting notions of induced diffusion or elastic
scattering at either location.
2. Instruments and data
Measurements were obtained near the Hawaiian Ridge
during two 6-week deployments of the research platform
FLIP as part of HOME. A map showing both locations
is shown in Fig. 1. During the 2002 HOME Nearfield
program, R/P FLIP was moored at 21.688N, 158.638W,
on the southwest shoulder of Kaena Ridge. This ridge,
formed by the northwest extension of the island of Oahu,
was one of the most energetic tidal conversion sites
identified during HOME. Shear levels 4 times greater
1526
JOURNAL OF PHYSICAL OCEANOGRAPHY
FIG. 1. Deployment of R/P FLIP in 2001–02 (Rainville and
Pinkel 2006a). As part of HOME, FLIP was moored at two locations near Kaena Ridge, Hawaii. During the 2002 Farfield component, the measurement location was approximately 430 km
to the southwest of the ridge crest, in the approximate path of an
M2 tidal beam (model fluxes are shown by the arrows). The 2002
Nearfield component placed FLIP on the shoulder of Kaena Ridge
in approximately 1100 m of water, in a location intersecting the
southward-propagating ray emanating from the north ridge.
and vertical displacements 10 times larger than openocean levels were reported near the ridge (Rudnick et al.
2003; Rainville and Pinkel 2006a). In the 2001 HOME
Farfield component, FLIP was deployed 430 km to the
south-southwest of Oahu, at 18.398N, 160.708W, in the
approximate path of an M2 tidal beam emanating from
Kaena Ridge. Here, wave energy levels were found to
be more representative of the open ocean (Rainville and
Pinkel 2006b).
The same instrumentation was used for both deployments (Fig. 2). A pair of Seabird SBE11 CTDs, sampling
at 24 Hz and profiling at approximately 3.6 m s21, combined to record temperature and salinity over a depth
range of 10–800 m once every 4 min with a vertical resolution of 2 m. Vertical velocities were inferred from the
VOLUME 42
FIG. 2. Schematic of R/P FLIP instrumentation during HOME,
2001–02. Tandem Seabird SBE11 CTDs profiled down to approximately 800 m once every 4 min with approximately 2-m resolution. The Deep-8 Doppler sonar recorded horizontal velocities
with approximately 4-m vertical resolution. Vertical velocities
were inferred from the motion of isopycnals as measured by the
CTDs. More than 11 000 profiles were collected during the Nearfield and more than 9000 were collected during the Farfield.
motion of isopycnal surfaces, as estimated from the CTD
record.
Horizontal velocities were measured by the Deep-8
Doppler sonar, which was deployed at 400-m depth
in the Nearfield and 320-m depth in the Farfield. The
Deep-8 was equipped with four upward-looking and
four downward-looking beams, transmitting repeatsequence coded pulses every 2 s and achieving a vertical resolution of 4 m (Pinkel and Smith 1992). The
reader is referred to Rainville and Pinkel (2006a) for a
more detailed discussion of the instruments and deployment.
Approximately 12 000 CTD profiles, spanning 33 days,
were recorded in the 2002 Nearfield component. Nearly
10 000 profiles were taken over a 28-day span during
the 2001 Farfield component.
SEPTEMBER 2012
SUN AND PINKEL
1527
FIG. 3. HOME Nearfield: depth–time maps. Example 3-day windows of the Nearfield data are shown, emphasizing
different spatial and temporal scales. Plots are WKB-scaled (4) and WKB-stretched (5). (top left) Isopycnal displacement h: long vertical scales and predominantly downward phase propagation with semidiurnal period are
dominant. (top right) Vertical velocity w: high-frequency motions with long vertical scales are most apparent here.
(bottom left) Meridional velocity y: a mix of semidiurnal and diurnal frequencies can be seen, with a greater mix of
vertical scales than in w. (bottom right) Vertical shear ›y/›z: much smaller vertical scales are apparent here than in
the other records. Shear layers are visibly deformed by the vertical heaving of the tides. The Deep-8 sonar was
deployed at 400 m (550 m in the WKB-stretched coordinate), resulting in a gap between 544- and 557-m depths in the
record. Bands in depths of 203–214 m and 645–666 m are caused by surface and bottom reflections, respectively, in
the sonar data. The gaps and discontinuities were interpolated in the velocities but are still apparent in the shears.
a. Depth–time maps
Example 3-day windows of isopycnal displacement,
horizontal and vertical velocities, and vertical shears
from each location are presented in Figs. 3 and 4. To allow
a better visual comparison between depths, Wentzel–
Kramers–Brillouin (WKB)-scaled vertical coordinates
(Gill 1982) have been used at each site. A cruise-averaged
profile of the buoyancy frequency N(z) defines the scaled
coordinate,
ð
1 z
N(z9) dz9, d,
(4)
zWKB 5
N0 0
where the constant N0 has been chosen so that zWKB spans
the same profiling range as the original coordinate z.
Velocities and shears have also been scaled,
1/2
N
,
w 5 w0
N0
21/2
N
(u, y) 5 (u0 , y 0 )
,
N0
21
N
,
(uz , yz ) 5 ((u0 )z , (y 0 )z )
N0
(5)
where each subscript 0 indicates the original quantity.
The scaling for linear propagation has been used for
velocities, but observed shears at both locations were
found to vary nearly as N rather than the predicted N 3/2 ,
so a scaling of (N/N0)21 is used for the shears.
Several vertical discontinuities exist in the sonar data
and are most visible in the shears. There is a gap around
the Deep-8, due to the separation between the upwardlooking and downward-looking sonar records, which
appears at depths of 544–557 m in the Nearfield and
483–502 m in the Farfield. Reflections from the sea surface contaminate depths of 203–214 m and 177–200 m,
respectively. A bottom reflection, which appears only
in the Nearfield, is also present in depths of 645–666 m.
To minimize the effect of these vertical discontinuities,
Eulerian coordinates are used through the present study.
This departs from the approach of Rainville and Pinkel
(2006a) and others, who have used semi-Lagrangian
(isopycnal) coordinates to reduce the effects of vertical
advection by the semidiurnal tide. In the conversion to
a semi-Lagrangian frame, the sonar artifacts are spread
across all isopycnals that pass through the affected
1528
JOURNAL OF PHYSICAL OCEANOGRAPHY
VOLUME 42
FIG. 4. HOME Farfield: depth–time maps. As in Fig. 3, but for the Farfield. (top left) Isopycnal displacement h:
long vertical scales are evident but no dominant sense of vertical propagation is seen. (top right) Vertical velocity w:
high-frequency motions with long vertical scales are seen as in the Nearfield. (bottom left) Meridional velocity y:
a mix of scales is seen, with less visible similarity to h. (bottom right) Vertical shear ›y/›z: as before, small vertical
scales with significant layer deformation are seen. Here the Deep-8 sonar was deployed at 320 m (490 m in the WKBstretched coordinate), resulting in a gap between 483- and 502-m depths in the record. The band in depths of
177–200 m is caused by a surface reflection in the sonar data. The gaps and discontinuities were interpolated in the
velocities but are still apparent in the shears.
depths. By remaining in the Eulerian frame, we limit the
influence of the artifacts. We have carried out a parallel,
semi-Lagrangian analysis and find that the results reported here are not significantly affected by the coordinate choice. For visual presentation, the Eulerian
velocity data have been interpolated across the affected
depths, but the interpolated depths are discarded whenever statistical averages are subsequently formed.
Not seen in the figures is a brief interruption to CTD
profiling that occurred in the Nearfield around day 280
of 2002, when one of the CTDs collided with the Deep-8.
Profiling resumed with a reduced, 700-m maximum depth.
The quantities from the Nearfield, shown in Fig. 3,
display different spatial and temporal scales. Displacements h (top left) exceed 100 m in the maximum and are
dominated by waves with long vertical scales and approximately semidiurnal period. A pronounced downward
phase propagation of wave crests indicates predominantly
upward energy propagation. These features are the signature of the semidiurnal internal tide emanating upward from the ridge, as discussed in Rainville and Pinkel
(2006a). A local maximum in displacement appears between 350 and 500 m in stretched coordinates and is
associated with a tidal beam that crosses the profiling
column near those depths.
Meridional velocities y (Fig. 3, bottom left), which are
approximately cross-ridge in the Nearfield, largely resemble the displacements. Peak magnitudes are approximately 25 cm s21. The period is still predominantly
semidiurnal, but a more complicated vertical structure
is visible in addition to the low-mode pattern seen in
the displacements.
Vertical velocities w (Fig. 3, top right) and vertical
shears y z (Fig. 3, bottom right) show opposite extremes
in scale. Only a hint of the semidiurnal wave crests in h
are visible in w. Instead, the record is filled by much
higher-frequency motions, with some motion at short
scales but much of it at long wavelengths. In contrast,
fine vertical scales on the order of 100 m and low frequencies dominate the shear record. Although hints of
semidiurnal and near-inertial phase variations can be
seen following shear layers, the most prominent visual
effect is that of vertical heaving of the shear layers associated with the semidiurnal tide.
The Farfield displacements h, presented in Fig. 4 (top
left), also have relatively long vertical scales as seen in
SEPTEMBER 2012
SUN AND PINKEL
1529
FIG. 5. HOME Nearfield: frequency–wavenumber spectra. Positive (negative) wavenumbers
denote upward- (downward-) propagating wave energy. (top) Vertical velocity w: most of the
variance is concentrated near M2 5 1.93 cpd and a set of apparent tidal harmonics, all of which
are predominantly associated with up-going wave energy. The spectrum gradually becomes
more symmetric with respect to vertical propagation at frequencies greater than about 12 cpd.
(middle) Complex horizontal velocity (u 1 iy): near-inertial and semidiurnal variance is apparent, along with the first few tidal harmonics. (bottom) Complex vertical shear (u 1 iy)z:
here, variance is concentrated in the near-inertial peak. More near-inertial energy is going
downward than upward. The K pattern that broadens with wavenumber is due in large part to
Doppler shifting of low-frequency waves by the vertical motions of the tides.
the Nearfield, but here the displacements show no clear
sense of vertical phase propagation. This is consistent
with the view that the internal tide gradually assumes
a low-mode structure as it propagates away from the
generation site. Rainville and Pinkel (2006a) reported
that nearly all the semidiurnal variance at the Farfield
site was contained in the first three modes, with more
than 60% in mode 1.
Velocities y in the Farfield (Fig. 4, bottom left) also
show a mix of longer vertical scales and shorterwavelength features, but unlike in the Nearfield there
is limited resemblance between h and y. In particular,
a series of shorter-scale motions visible below 300 m in
the Farfield have much smaller scales than seen here
above 300 m or in any part of the water column in the
Nearfield.
Vertical shears in the Farfield (Fig. 4, bottom right)
show finer vertical scales and a somewhat more coherent
wave group structure than in the Nearfield. It is difficult
to visually identify a dominant oscillation period. As
before, vertical velocities w (Fig. 4, top right) show a
slight hint of the structures seen in h embedded in a field
of high-frequency motions.
b. Spectral estimates
The full depth–time records of WKB stretched and
scaled data are two-dimensional (2D) Fourier transformed after applying a 5% Tukey taper in both depth
and time to reduce spectral leakage. Raw variances are
smoothed by a 3 bin 3 5 bin moving average, corresponding to approximately 0.003 cpm in vertical wavenumber 3 0.1 cpd in frequency, to produce the vertical
wavenumber–frequency (k, v) power spectra shown in
Figs. 5 and 6. Horizontal velocities were combined in
complex form u 1 iy to produce rotary power spectra.
As a result of our sign convention, spectral variance
in quadrants II and IV (where frequency and vertical wavenumber have opposite sign) corresponds to
1530
JOURNAL OF PHYSICAL OCEANOGRAPHY
VOLUME 42
FIG. 6. HOME Farfield: frequency–wavenumber spectra. As in Fig. 5, but for the Farfield.
Positive (negative) wavenumbers denote upward- (downward-) propagating wave energy. (top)
Vertical velocity w: only the semidiurnal tidal frequency and its first harmonic are clearly
identifiable. Vertical energy propagation is roughly symmetric. (Middle) Complex horizontal
velocity (u 1 iy): near-inertial and semidiurnal variance is apparent here. (Bottom) Complex
vertical shear (u 1 iy)z: variance is concentrated in the near-inertial peak, with roughly equal
parts going upward and downward. The K pattern that is due in large part to Doppler shifting of
low-frequency waves by the tides is visible here as well.
upward-propagating wave energy, while quadrants I
and III (where frequency vertical wavenumber and frequency have the same sign) show downward-propagating
wave energy.
In the Nearfield spectra, shown in Fig. 5, vertical velocities (Fig. 5, top) show a progression of vertical ridges
that are nearly symmetric about the origin and associated with predominantly upward energy propagation.
The strongest ridges are associated with the low-mode
internal tide at the principal lunar semidiurnal frequency,
6M2 5 61.93 cpd, corresponding to a period of 12.42 h.
Harmonics of the tide are also evident at 62M2, 63M2,
64M2, . . . The distinction between M2 and the solar
semidiurnal frequency, S2 5 2.0 cpd, is apparent by the
fourth harmonic of M2, at 9.66 cpd. A series of lesser
ridges can be seen between the M2 harmonics and may
be Doppler-shifted images of the M2 harmonics. The
extension of the ridges at nM2 frequency to small downward wavenumbers may be due in part to spectral
leakage from the upward half plane. The asymmetry
gradually subsides with increasing frequency, so that
waves above 12 cpd have roughly equal upward and
downward variance. Outside of the large peaks, variance
appears relatively constant with increasing frequency.
The largest horizontal velocity variances are found
near 6M2 frequency, again at small wavenumbers and
with a distinctly upward sense of propagation. Variance
at near-inertial frequencies rolls off more slowly with
wavenumber than does the tide and is associated with
both upward and downward waves. The downward nearinertial ridge contains somewhat more variance and
peaks at slightly lower frequency than the upward
ridge. A smaller ridge is also apparent at approximately
2(1/2)M2 frequency and is concentrated at somewhat
smaller wavenumbers than the near-inertial variance.
A downward ridge of unknown origin is seen at about
23 cpd. Overall, the horizontal velocity spectra roll off
quickly with frequency, and in the more apparent lower
frequencies, somewhat more upward than downward
energy is seen as in the power spectra of w.
SEPTEMBER 2012
SUN AND PINKEL
1531
FIG. 7. Frequency spectra of vertical velocity. The top spectrum is Nearfield w: excess
upward-propagating wave energy is visible at the semidiurnal frequency M2, a series of
harmonics, and at the subharmonic (1/2)M2 . The bottom spectrum is Farfield w: the Farfield
spectrum has been shifted down by one order of magnitude. There is little if any vertical
asymmetry visible in the Farfield.
Shear variance is concentrated at a near-inertial peak,
with wavenumbers ’60.01 cpm associated with downward propagation. Significant but lesser variance is also
associated with upward near-inertial propagation. The
peak at 2(1/2)M2 frequency is more distinct in the shear
spectra. Note the expanded wavenumber limits in the
shear plot relative to the velocity spectra to accommodate the decreased slope with increasing wavenumber.
Despite the change in slope, shears still fall off above
about 0.025 cpm. A widening pattern of shear variance
is apparent with increasing vertical wavenumber. This
‘‘hourglass’’ shape is associated with Doppler shifting in
these Eulerian wavenumber–frequency spectra (Pinkel
2008).
In contrast to the Nearfield, the Farfield spectra
are not noticeably asymmetric with respect to vertical propagation. Farfield vertical velocity variance
(Fig. 6, top) is confined to small wavenumbers as in the
Nearfield. Distinct lines appear at frequencies 6M2
and 62M2, but 63M2 and 64M2 are difficult to distinguish from the background. Horizontal velocity
spectra also show a low-mode M2 tide. A better separation between f and (1/2)M2 is seen in the Farfield
because of the slightly lower local inertial frequency.
Shear variance is mainly distributed near the inertial
frequency.
Frequency spectra are obtained by integrating the
spectral variance in Figs. 5 and 6 over all vertical wavenumbers. Vertical velocity spectra from both the Nearfield and Farfield are presented together for comparison
in Fig. 7. Upward- and downward-traveling energy are
overplotted on the same log–log axes. The Farfield
spectra (lower plots in each panel) have been shifted
downward by one order of magnitude for visual clarity.
The ridges in upward energy in the Nearfield are visible
here as sharp peaks in the upper plots, with the upward
semidiurnal variance about 5 times as large as the downward variance. The Nearfield variances remain relatively
steady with rising frequency, with a hint of a rolloff
before 102 cpd, whereas the Farfield variances appear
to rise slightly before beginning a rolloff at slightly
lower frequency. A sharp rolloff is not seen in either
plot because the spectra represent an average of frequency spectra from depths across which N varies by
a factor of 3–4.
Horizontal velocities and vertical shears (Fig. 8) tell
much the same story. The Nearfield semidiurnal velocity
peak is nearly an order of magnitude larger in upward
variance than in downward variance, and the upward
tidal harmonics are distinct until about 10 cpd. However, downward near-inertial shear variance is somewhat larger than upward near-inertial shear variance.
The Farfield spectra, by contrast, are quite symmetric
in all respects.
Vertical wavenumber spectra, shown in Fig. 9, are
obtained by integrating the wavenumber–frequency spectra across all frequencies. Our ability to resolve vertical
wavenumbers is limited relative to our ability to resolve
frequencies. Low wavenumber resolution is limited by
the profiling range, which was about 600 m in the vertical. Vertical velocity variance, shown in the left panel
of Fig. 9, is concentrated at the two lowest wavenumbers that could be resolved, with a falloff of approximately m23/2 in the Nearfield and a somewhat
1532
JOURNAL OF PHYSICAL OCEANOGRAPHY
VOLUME 42
FIG. 8. Frequency spectra of horizontal velocity and vertical shear. As before, the Farfield spectra have been
shifted down by one order of magnitude. (Left) Complex horizontal velocity (u 1 iy): the series of tidal harmonic
peaks are visible in the Nearfield, with excess up-going energy. In the Farfield, the near-inertial and diurnal peaks are
resolved. The semidiurnal peak is accompanied by only one or two significant harmonics. (right) Complex vertical
shear (u 1 iy)z: here, similar features are seen as in the velocity spectra, but the near-inertial peak is greatly emphasized.
Only near-inertial shears are markedly unequal in the Nearfield, with more wave energy propagating downward.
steeper m25/2 in the Farfield. The wavenumber spectra
are much smoother than the frequency spectra because
of the large number of profiles that have been averaged.
The slope steepens in the Nearfield above wavenumbers
of about 3 3 1022 cpm. The Farfield variance flattens
out briefly around 1 3 1022 cpm before falling off more
steeply again above 7 3 1022 cpm.
The vertical shear wavenumber spectra, shown on the
right panel of Fig. 9, are relatively similar between the
two sites. The Farfield shear variance peaks at about
1/ 60 cpm, whereas variance in the Nearfield peaks at a
slightly lower wavenumber, about 1/ 90 cpm. Shear variance falls off on both sides of the peak wavenumber in
all the spectra except the Nearfield downward, where
the peak is broadened into a plateau between about 1/ 200
and 1/ 90 cpm.
3. Stress–shear triple correlation
Products of the form hu9i u9j i›Ui /›xj , where ›Ui/›xj
represents a mean strain rate and the hu9i u9j i are timeaveraged Reynolds stresses, are familiar as the shear
production term from turbulence theory (Tennekes and
Lumley 1972). Studies of wave–mean flow interactions
have also examined analogous correlations, in which the
stresses are presumed to be due to wave activity rather
than turbulent motions and the mean strain rates are
assumed to be primarily vertical shears (Ruddick and
Joyce 1979; Gargett and Holloway 1984; Jacobs and Cox
1987; Duda and Jacobs 1998).
Our method for demonstrating energy transfers between the energetic low-frequency shears and higherfrequency waves focuses on triple correlations of the form
SEPTEMBER 2012
1533
SUN AND PINKEL
FIG. 9. Wavenumber spectra of vertical velocity and shear. As before, the Farfield spectra have been shifted
down by one order of magnitude. (Left) Vertical velocity w: variance is concentrated in the lowest few wavenumber bins. The Farfield velocity rolls off more steeply with vertical wavenumber before about 1022 cpm, relative to the Nearfield. (bottom) Vertical shear (u 1 iy)z: the Farfield spectra are sharply peaked at 1/ 60 cpm.
The Nearfield peak in up-going energy is at 1/ 90 cpm. Down-going energy in the Nearfield is extended in a band
between 1/ 200 and 1/ 90 cpm. Spectra have not been corrected for the 4-m centered differencing used to calculate
shears.
›U
u9i w9 i ,
›z
i, j 5 1, 2,
(6)
where as before the angle brackets hi indicate time
averaging. Unlike in the turbulent production term, the
›Ui/›z in (6) are not mean shears but instead oscillate
at low frequency. Meanwhile, the ‘‘fluctuation’’ stresses
u9i w9 are associated with motions at higher frequencies
than the ›Ui/›z and are assumed, but not required, to be
internal waves. We differ from Duda and Jacobs (1998)
and others by not explicitly assuming a vertical scale
separation between the high-frequency waves and the
low-frequency shears.
To relate the triple product (6) to an energy transfer
between low- and high-frequency waves, it is useful to
return to the equations of motion,
›u
1 u $u 1 f 3 u 5 2$p/r0 2 $F,
›t
(7)
where u 5 (u, y, w) is the velocity vector in Cartesian
components, f is the Coriolis frequency, p is pressure,
F is the geopotential, and r0 is the density profile at
rest.
We proceed by writing the total flow as a sum of lowand high-frequency parts,
u 5 U 1 u9,
p 5 p~ 1 p9,
r 5 r~ 1 r9.
(8)
An energy evolution equation for the high-frequency
fluctuations is obtained by taking the scalar product of u9
with (7) and time averaging over a period that is long
compared to the time scales of both u9 and U,
1534
JOURNAL OF PHYSICAL OCEANOGRAPHY
1›
›U
ku9k2 1 u9 1 hu9 [(U 1 u9) $(U 1 u9)]i
2 ›t
›t
1
^ ,
r 1 r9)k]
(9)
5 2u9 [$( p~ 1 p9) 1 g(~
r0
where ku9k2 5 u9 u9 is the Euclidean norm.
Assuming a statistically steady state and no overlap
in frequency between U and u, the correlation between a
low-frequency and a high-frequency quantity must vanish,
›U
u9 5 0, h$~
p u9i 5 0, h~
rw9i 5 0.
›t
(10)
Next we expand the triple correlations, keeping only quadratic terms in the primed (high-frequency) quantities,
hu9 [(U 1 u9) $(U 1 u9)]i
5 hu9 (U $U)i 1 hu9 (U $u9)i 1 hu9 (u9 $U)i.
(11)
The first correlation on the right-hand side, hu9 (U $U)i,
is formed by the product of one high-frequency and two
low-frequency quantities. Triads interactions between
these frequencies include PSI, particularly PSI involving
semidiurnal tides and pairs of near-diurnal subharmonics.
Such an interaction in HOME Nearfield is the subject
of a companion paper (Sun and Pinkel 2012, manuscript
submitted to J. Phys. Oceanogr.) and will not be addressed here. In this analysis, we avoid the influence of
PSI by requiring a frequency gap of at least one octave
between the low- and high-frequency fields. Any pair of
low-frequency waves would be frequency resonant with
a third wave whose frequency is too low to be found in the
high-frequency field. The triple correlation hu9 (U $U)i
is thus zero. The second term, hu9 (U $u9)i 5
U $(1/2)u92 , represents the advection of high-frequency
kinetic energy by the low-frequency motions, which, unlike in the wave–mean flow problem, are assumed to be
oscillatory, so that the net advection is zero. The remaining term, hu9 (u9 $U)i 5 hu9i u9j ›Ui /›xj i, is the triple correlation of interest that appears in (6).
Thus, using (10) and retaining the nonzero triple correlation in (11), the time-averaged energy equation (9)
becomes
›1
2
ku9k 1 hu9 (u9 $U)i
›t 2
1
g
r9w9 .
(12)
5 2 $ ( p9u9) 2
r0
r0
The first and last terms have the familiar forms of the
time derivatives of averaged kinetic and potential energy densities,
›hE9k i
5
›t
›1
ku9k2 ,
›t 2
VOLUME 42
›hE9p i
›t
5
1
hgr9w9i,
r0
(13)
whose sum we define as the total averaged energy
density,
hE9i 5 hE9k i 1 hE9p i.
(14)
Thus, using index notation for the triple correlation, (12)
can be expressed as
*
+
›Ui
›hE9i
1
5 2 u9i u9j
2 $ h p9u9i,
›t
r0
›xj
(15)
which represents an energy balance between the highfrequency source 2u9i u9j ›Ui /›xj and the divergence of
energy flux due to high-frequency waves. For u9 composed of a high-frequency wave packet with energy
density E9 and group velocity c9g , the energy flux is
1
( p9u9Þ 5 c9g E9,
r0
(16)
so that (15) has the interpretation
*
+
›Ui
›hE9i
,
1 $ c9g hE9i 5 2 u9i u9j
›t
›xj
(17)
in which h2u9i u9j ›Ui /›xj i balances the change in average
energy hE9i in a frame moving with the group velocity.
Generally, the low-frequency vertical shears are
thought to dominate over the horizontal shears. Unfortunately, we have no direct way of evaluating the
correlations h2w9u9i ›W/›xi i involving horizontal stresses
and horizontal velocity gradients. Horizontal homogeneity
of the internal wave field is sometimes invoked to justify
dropping the horizontal stress–shear terms (Gargett and
Holloway 1984), but here such an assumption may be
problematic because the wave field in the Nearfield is
characterized by horizontally inhomogeneous features
such as tidal beams. Keeping these difficulties in mind,
we proceed by retaining only the correlations involving low-frequency vertical shears and assign them the
symbol
›U
(18)
«* 5 2 u9i w9 i , i 5 1, 2,
›z
to emphasize the analogy between our high-frequency
source term and the turbulent shear production term.
An associated correlation coefficient r*, which measures
the relative coupling between the waves contributing to
SEPTEMBER 2012
1535
SUN AND PINKEL
FIG. 10. Nearfield: stress–shear triple product. (left) Triple product (2uiw›U/›z): red patches indicate positive
energy transfer from low-frequency shears to high-frequency waves. Most of the activity occurs between 400- and
600-m depth, where the shear maximum is located and also roughly where a tidal beam crosses the profiling column.
(right) Energy transfer rate «*: the blue line is a profile of «*, the time-averaged energy transfer rate. The red line «err
is an error estimate calculated using the unaligned horizontal terms.
«* independent of their respective energy levels, can
be defined by normalizing «* by the root-mean-square
(RMS) magnitude,
«*
ffi.
r* 5 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
h(u9i ) ih(w9)2 ih(›Ui /›z)2 i
(19)
Estimates of «* in HOME
Depth–time maps of high-frequency u9i and w9 and
low-frequency (›Ui/›z) in both HOME locations are
formed from WKB stretched and scaled quantities. To
obtain (›Ui/›z), the vertical shears are filtered in frequency to allow a passband of 1/ 40–1/ 8 cph, or 0.6–3 cpd,
which includes the most of the shear in the inertial, diurnal, and semidiurnal bands as seen in the spectra of
Fig. 8. Meanwhile, w and ui are high-pass filtered above
6 cpd, leaving a frequency gap to exclude PSI-like
interactions. All filters are zero-phase, sixth-order
Butterworth. To form the triple products that appear
inside the brackets in (18), the bandpassed u9i , w9, and
(›Ui/›z) are multiplied together pointwise and summed over the index i.
Depth–time maps of 2u9i w9(›Ui /›z) are shown in
Figs. 10 and 11. The plots have been smoothed by 10 m
in depth and 1 h in time. A patch of predominantly red
color is apparent in the Nearfield triple product (Fig. 10)
below 300-m depth on days 260–270. The relatively
uniform color indicates sustained interactions resulting
in a positive net energy transfer from low-frequency
shears to high-frequency waves. The patch is strongest
between spring semidiurnal tide and the maximum of
near-inertial variance around day 270. Between days
276 and 287, more red streaks and patches are visible
below 300 m, including a deeper patch below 550 m on
days 281–286. There is a band of relative quiet between
approximately 150- and 300-m depths, part of which
is due to the surface reflection in the sonar data near
200 m, above which there are more active interactions.
The profile of «*, plotted to the right of the main plot,
is a time average of triple products. Physical units of
energy transfer rate per unit mass are preserved by
computing averages from raw, rather than WKB-scaled,
quantities. The energy transfer rate in the Nearfield, calculated as a simple average over depth (non–WKB scaled)
using unscaled quantities, is about 2 3 1027 W kg21, with
a maximum value of 2.8 3 1027 W kg21 near a depth of
630 m. The correlation r* (not shown) has a similar
profile to «* but scaled by a constant, with a depthaveraged r* of 0.075.
Measurement error is a concern in the energy transfer
estimates. To provide an intuitive estimate of the error,
we compute a test correlation «err from velocities and
shears that have the same magnitudes as in «* but are
unaligned and thus are not physically meaningful,
›Uj
,
«err 5 2 u9i w9
›z
i, j 5 1, 2, i 6¼ j.
(20)
The error estimates «err, plotted on the same axes as
the «* profile, are found to be smaller than « by about
an order of magnitude.
1536
JOURNAL OF PHYSICAL OCEANOGRAPHY
VOLUME 42
FIG. 11. Farfield: stress–shear triple correlation. As in Fig. 10, but for the Farfield. (left) Triple product (2uiw›Ui/›z):
red patches, indicating positive energy transfer from low-frequency shears to high-frequency waves, are more sparse
here than in the Nearfield. The burst of energy after day 300 may be associated with an eddy that passed through
the measurement site. (right) Energy transfer rate « : the blue line is a profile of « , the time-averaged energy
*
*
transfer rate. The green line is «* computed using only data prior to day 300. The red line «err is an error estimate
calculated using the unaligned horizontal terms.
Triple products in the Farfield (Fig. 11) are somewhat
smaller and patchier than in the Nearfield. Smaller
magnitudes are to be expected, because the product
of mean variances appearing under the square root in
the denominator of (19) was a factor of 5 smaller in the
Farfield. Average «* is about 8.8 3 1028 W kg21 with
a maximum of 1.9 3 1027 W kg21. Unlike in the Nearfield, this maximum appears near the top of the profile.
However, a second maximum, which is nearly as large,
at 1.8 3 1027 W kg21, appears near 430-m depth. Correlations were slightly higher than in the Nearfield, with
an average r* of 0.085. A strong patch of red interactions
intermixed with blue can be seen at middepth after day
300. This burst of activity may be associated with an
eddy that moved into the measurement area near the
end of the deployment. An alternate average, indicated
by the green line, was formed to assess the effect of the
burst by using only data before day 300. The differences
appear modest. «* decreased by about a third in a range
of about 80 m above and below 430 m, and the local
maximum dropped to 1.1 3 1027 W kg21 using the alternative averaging period. However, the maximum near
the top of the profile increased to 2.3 3 1027 W kg21.
4. Energy transfer bispectra
Although «* quantifies the net energy transfer between low- and high-frequency waves, bispectral methods
(Hinich and Clay 1968; Kim and Powers 1979) identify
specific frequencies and wavenumbers of waves that are
interacting. Bispectra have been used to examine nonlinear coupling in a variety of settings, including plasma
flows (Kim and Powers 1979) surface gravity waves
(Elgar and Guza 1985; Elgar et al. 1995), and internal
waves (McComas and Briscoe 1980; Furue 2003; Furuichi
et al. 2005; Carter and Gregg 2006; Frajka-Williams et al.
2006; Sun and Pinkel 2012, manuscript submitted to
J. Phys. Oceanogr.).
Consider three random, real variables x, y, z with
Fourier coefficients in frequency–wavenumber (v, k)
space that are denoted by X(v, k), Y(v, k), Z(v, k).
The bispectrum of these variables is defined as
B(v1 , k1 , v2 , k2 )
5 EbX(v1 , k1 ), Y(v2 , k2 ), Z(2v1 2 v2 , 2k1 2 k2 )c,
5 E[X(v1 , k1 ), Y(v2 , k2 ), Z*(v1 1 v2 , k1 1 k2 )],
(21)
where E[] indicates an expected value.
In three spatial dimensions, k 5 (k, l, m), the bispectrum is a function defined over an eight-dimensional
space. Here we consider only vertical wavenumbers,
so that four-dimensional bispectra are estimated. The
analysis is further simplified if we consider bispectra of
either wavenumber or frequency alone. Because the
energetic shears in the observations are concentrated
at low frequencies and we are primarily interested in
SEPTEMBER 2012
1537
SUN AND PINKEL
identifying up-frequency transfers, we begin by examining frequency bispectra. We define the frequency bispectrum over Fourier frequency pairs (v1, v2) as
B(v1 , v2 ) 5 E[X(v1 ), Y(v2 ), Z*(v1 1 v2 )].
(22)
The magnitude of the bispectrum depends on both
on the magnitudes and relative phases of the respective
Fourier coefficients. The normalized bispectrum, also
called the bicoherence, measures phase locking only.
We use a normalization (McComas and Briscoe 1980;
Elgar and Guza 1988; Hinich and Wolinsky 2005), which
has a form consistent with (19), for the bicoherence,
5
å å U j,m W n Z*j,m1n .
m
(30)
n
The expected values of the terms in the summation are
exactly the bispectrum (22) between ui, w, and ›Ui/›z at
each (vm, vn). Their sum is a real-valued energy transfer, so that we can restrict one of m or n to positive
values while keeping the summands real-valued by
combining complex conjugates,
N /221
N /221
å
å
m52N /2 n52N /2
U j,m W n Z*
j,m1n
N /221 N /221
jB(vm , vn )j
.
b 5 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
E[jX(v1 )j ]E[jY(v2 )j2 ]E[jZ*(v1 1 v1 )j2 ]
(23)
respectively.
The direct relationship between the bispectrum (22)
and «* can be shown by expressing the quantities uj, w,
and ›Uj/›z as Fourier series with N coefficients,
›Ui /›z 5
å U j,n eiv t ,
n
w9(t) 5
n
å Z j,n e
ivn t
å W n eiv t ,
,
m
5
t
n
(25)
n
p
m
1vn 1vp )t
.
(27)
p
Time averaging over the length of the Fourier transform causes all the exponential terms except those with
zero frequency to vanish; thus,
hu9j w9›Uj /›zi 5
å å å U j,m W n Z j,p d(vm 1 vn 1 vp ),
m
n
p
(28)
5
å å U j,m W n Z j,2m2n ,
m
or equivalently
n
U j,m W n Z*
j,m1n 1 U*
j,m W*
n Z j,m1n
(31)
å
2 <fU j,m W n Z*j,m1n g.
(32)
We define the energy transfer bispectrum between u9i ,
w9, and ›Ui/›z as the expected values of terms appearing
in the sum, where we have included both horizontal directions and incorporated a sign change such that a
positive energy transfer is from low to high frequency,
B* (vm , vn ) 5 E[22 <fU 1,m W n Z*1,m1n
1 U 2,m W n Z*2,m1n g].
(33)
By this definition,
N / 221 N / 221
å W n eivn t å Z j,p eivp t (26)
å å å U j,m W n Z j,p ei(v
m
å
m51 n52N /2
«* 5
vn 5 2pn/N,
Then
m
5
n
n
å U j,m eiv
å
n
N
N
n 5 2 , . . . , 2 1.
2
2
u9j w9›Uj /›z 5
å
m51 n52N /2
N /221 N /221
Statistically significant phase locking is demonstrated
when the sample bicoherence exceeds zero by an appropriate threshold, where thresholds for 90%, 95%, and 99%
confidence at m degrees of freedom (DOF) are given by
pffiffiffiffiffiffiffiffiffiffiffiffi
pffiffiffiffiffiffiffiffiffiffiffiffi
b90% 5 4:6/m, b95% 5 6:0/m, and
pffiffiffiffiffiffiffiffiffiffiffiffi
(24)
b99% 5 9:2/m,
u9j (t) 5
5
(29)
å
å
m51 n52N /2
B* (vm , vn ).
(34)
a. Frequency bispectra
The original depth–time records of u9i , w9, and ›Ui/›z,
rather than the WKB-scaled versions, are used to form
frequency bispectra. This allows direct comparison between energy transfer bispectra and the energy transfer rates «*, shown in Figs. 10 and 11, which are also
computed using unscaled data. Time series from each
Eulerian depth are Fourier transformed in nonoverlapping segments of length 1024, or about 68.3 h. A 5%
Tukey taper is applied to each segment. Separate bispectral realizations are formed for each depth–time
segment between 100 and 700 m depth. The realizations, excluding the contaminated sonar depths, are
then averaged together to form sample estimates for the
bispectrum and bicoherence. Both B* and the squared
bicoherence b2 are smoothed with a 3 3 3 bin moving
average, which spans an approximately 1 cpd 3 1 cpd
region. The resulting bispectral densities are presented
in Figs. 12 and 13. Each value B*(vu, vw) is plotted on
1538
JOURNAL OF PHYSICAL OCEANOGRAPHY
VOLUME 42
FIG. 12. Nearfield: energy transfer bispectral estimates. (left) Energy transfer bispectrum B*(vu, vw): bispectral
densities plotted at coordinates (vu, vw) correspond to triads (vu, vw, 2vu, 2vw). Most of the positive energy
transfers (red) are concentrated near the diagonal line vu 5 2vw, but some negative energy transfers (blue) are seen
at low frequencies as well. (right) Bicoherence b(vu, vw): the color scale is adjusted so that white and red represent
statistically significant values larger than b95% 5 0.019.
the bifrequency plane at the coordinates (vu, vw). Following the indexing convention in the summation (33),
the horizontal axis is restricted to positive frequencies
vu, whereas frequencies vw on the vertical axis are
signed to account for both sum and difference interactions. The third frequency, which is associated with the
Fourier coefficients of the shears (›Ui/›z), is always the
negative sum of the coordinates, (v 5 2vu 2 vw). It will
sometimes be useful in the discussion to indicate the
third frequency explicitly. For example, the bispectral
density plotted at the coordinates (16, 217) cpd is referred to explicitly as B*(16, 217, 1) cpd. No significant
bispectral values are found in the lower half plane,
where interactions would involve vu and vw with the
same signs and a high-frequency shear, so only the
upper half plane is shown in the figures.
Positive energy transfers, shown in red in the energy
transfer bispectra, are spread along the diagonal vw 5
2vu in both Figs. 12 and 13, suggesting that pairs of
high-frequency waves are interacting with and receiving
energy from shears with close to zero frequency. Negative energy transfers, colored blue, also appear in both
figures near the vertical axis vu 5 0. In addition, a pair
of positive ridges is seen only the Nearfield bispectrum
(Fig. 12) near vu 5 2 cpd, vw 5 22 cpd and may be
associated with PSI energy transfers.
From (24) we estimate a 95% confidence threshold for
zero bicoherence of b95% 5 0.042 in the Nearfield and
b95% 5 0.044 in the Farfield. The color scale for the bicoherence plots is chosen so that values at the threshold
appear in white, with larger values tending toward red.
In estimating the effective degrees of freedom in the
bicoherence estimate, we regard measurements separated by 50 m, or approximately half a near-inertial
wavelength, as independent. This results in 186 independent realizations for each of two horizontal directions in the Nearfield and 174 realizations in the Farfield.
Both estimates are improved by a factor of 9 by the 3 3 3
smoothing in bifrequency space.
The significant bicoherence in Figs. 12 and 13 is, like
the positive bispectra, concentrated along the diagonal
(vw 5 2vu). However, unlike the bispectra, the bicoherence appears in a band that widens with increasing
frequency. Notably, bicoherences are also spread at isolated points throughout the bifrequency plane, especially
at frequencies above 48 cpd in Fig. 13. The increasing
spread of bicoherences at high frequencies may be due
to Doppler shifting or measurement noise that has contaminated the Fourier coefficients. Fortunately, the noise
seems to appear mostly where the bispectra are small,
so that the energy transfer estimates are relatively
unaffected.
SEPTEMBER 2012
SUN AND PINKEL
1539
FIG. 13. Farfield: energy transfer bispectral estimates. As in Fig. 12, but for the Farfield. (left) Energy transfer
bispectrum B (vu, vw): bispectral densities plotted at coordinates (vu, vw) correspond to triads (vu, vw, 2vu, 2vw).
*
Most of the positive energy transfers (red) are concentrated near the diagonal line vu 5 vw, but again some negative
energy transfers (blue) are seen at low frequencies. (right) Bicoherence b(vu, vw): the color scale is adjusted so that
white and red represent statistically significant values larger than b95% 5 0.020.
The bispectral energy transfers and bicoherence thresholds are combined in the bispectral plots of Figs. 14
and 15. Here, B*(vu, vw) has been masked to show
only significantly bicoherent energy transfers. Most of
the positive energy transfers near the diagonal remain
intact, whereas the negative energy transfers near vu 5
0 have been removed. A small remnant of the ridge
along vw 5 22 cpd also remains in Fig. 14.
The energy transfers near (vw 5 2vu) do not lie directly on the diagonal itself but rather along a pair of
slightly offset ridges. A detailed view of this feature is
provided in the inset in each figure. Guide lines drawn
along constant sum frequencies vu 1 vw 5 6f, indicate
where interactions between two high-frequency waves
and a pure inertial wave (frequency 6f ) would lie in the
plot.
Below the bispectra in Figs. 14 and 15 are plots showing net energy transfer due to high-frequency waves up
to and including the frequency v,
ðv
6 cpd
ðv
6 cpd
B* (vu , vw ) dvw dvu .
(35)
The integration begins at vu 5 6 cpd to be consistent
with «*, which was calculated with a gap between 3 and
6 cpd. The ridge in the Nearfield at vw 5 22 cpd falls
within the gap and is thus omitted from the integral.
There is a falloff in energy transfer with increasing
frequency beginning at about 72 cpd, with very little
contribution from frequencies above 96 cpd, which
exceeds N(z) at most depths. In the Nearfield, B* approaches 2.8 3 1027 W kg21. The Farfield value is
1.2 3 1027 W kg21. The estimates of energy transfer
formed by integrating the bispectra are in close agreement with the depth-averaged values of «* (Figs. 10, 11).
b. Wavenumber bispectra
Although the significant bicoherences in both the
Nearfield and the Farfield imply that interactions are
occurring between the energetic low-frequency shears
and a variety of high-frequency waves, the frequency
analysis gives insufficient information for determining
whether the interactions resemble ID, ES, or neither.
It should be possible to distinguish between the highfrequency members in ID (similar vertical wavenumbers
that are much larger than the low-frequency shear wavenumbers) and ES (similar but opposite-signed wavenumbers that are smaller than the shear wavenumbers).
A similar procedure as used in the frequency analysis
can be used to form vertical wavenumber bispectra.
WKB-scaled and WKB-stretched quantities are used
for the wavenumber analysis to minimize effects from
wave refraction. Although before B* was defined over
bifrequencies (vu, vw), here we define B*(ku, kw) over
1540
JOURNAL OF PHYSICAL OCEANOGRAPHY
FIG. 14. Nearfield: energy transfer bispectrum B*(vu, vw). As in
Fig. 12, but showing only bicoherent energy transfers. (top) A
detailed view inset shows that the sum frequencies of triads near
the diagonal are not precisely zero but differ by about f. (bottom) A
plot shows of the cumulative integral of energy transfer rate with
frequency.
vertical wavenumber pairs. Prior to Fourier transforming in wavenumber, the frequency separation between
low- and high-frequency fields is ensured by bandpass
filtering in the time domain, as was done in estimating
«*. The data are then windowed using a 5% Tukey taper
in depth. Only profiles 500–7499 are used from each
dataset. This particular span is chosen for three reasons: to allow time for the low-pass filter output to stabilize after the beginning of the record, to avoid the
interruption around day 8099 in the Nearfield, and to
exclude the eddy-associated activity after day 300 in the
Farfield. The number of profiles used from each site is
kept the same to simplify statistical comparison.
Vertical wavenumber bispectral estimates from the
Nearfield and Farfield are presented in Figs. 16 and 17.
Although only difference reactions are possible in the
VOLUME 42
FIG. 15. Farfield: energy transfer bispectrum B (vu, vw). As in
*
Fig. 13, but showing only bicoherent energy transfers. (top) A
detailed view inset shows that the sum frequencies of triads near
the diagonal are not precisely zero but differ by about f. (bottom)
A plot shows of the cumulative integral of energy transfer rate with
frequency.
bifrequency plane because of the frequency gap between
low- and high-frequency waves, no similar restriction
has been placed upon the interacting wavenumbers.
Both sum and difference interactions (kw both positive
and negative) are found to be important, and thus both
half planes are shown. The bispectral smoothing window, formed by convolving a 2 3 2 bin square filter with
itself, has been chosen because it has the same effective
width as the 2 3 2 bin square filter but is symmetric
around its center.
Nearfield energy transfers, shown in the left panel of
Fig. 16, are confined in a small region of biwavenumber
space that corresponds to small vertical wavenumbers
(ku, kw). In the Farfield (Fig. 17), the region is significantly elongated in the horizontal direction but remains
close to the horizontal ku axis. Bicoherence thresholds
SEPTEMBER 2012
SUN AND PINKEL
1541
FIG. 16. Nearfield: wavenumber bispectral estimates. (left) Wavenumber bispectrum B (ku, kw): nearly all the energy transfers are
*
concentrated at very low wavenumbers. (middle) Bicoherence b(vu, vw): the energy transfers appear to be bicoherent. Possible noise at
high frequencies, not associated with significant energy transfers, is apparent. (right) B*(ku, kw, 2ku, 2kw) showing bicoherent values
only: a contour plot is used to help locate the peak wavenumbers. Guide lines indicate constant sum wavenumbers (2ku, 2kw).
computed according to (24) are shown in the second
panel of each figure. Profiles separated in time by 2 h, or
a half period of the 6-cpd cutoff for the high-frequency
band, are taken as independent measurements. In both
HOME locations, the bicoherent regions contain the
regions with the largest energy transfer but also extend
in a band along the ku axis toward high wavenumber.
The rightmost panel of each figure shows a contour
plot of the low-wavenumber region that contains the
bicoherent energy transfers. A black tile indicates the
size of each (ku, kw) bin. Individual peaks are best seen
in the contour plots. The strongest peak in the Nearfield
is at on the sum interaction side, seen in Fig. 16 at (kw, ku,
2kw 2 ku) ’ (0.9, 0.2, 21.1) 3 1022 cpm, and involving
high-frequency waves with a vertical wavenumber ratio
ku /kw ’ 4:5/1, which is far from the approximately equal
wavenumbers expected for either ID or ES. A second
peak at (0.4, 0.2, 20.6) 3 1022 cpm has somewhat less
disparate ratio of about 2/ 1. However, the narrower
peak at (1.2, 20.2, 21.0) 3 1022 cpm is even further
from unity, at about 6/ 1. In the main interactions seen
here, kw is consistently in the second lowest wavenumber bin, which corresponds to a wavelength on the
order of 500 m.
In the Farfield, the high-frequency vertical wavenumbers tend to be even more widely disparate. Figure 17
shows a set of major peaks corresponding to sum interactions between waves at kw 5 0.15 3 1022 cpm and
waves at ku 5 1.5, 0.6, and 1.0 3 1022 cpm, where the
frequencies are listed in descending strength of the
interactions. On the difference side, waves with similarly long wavelength, at kw 5 20.15 3 1022 cpm, interact with waves at ku 5 1.9, 0.8, and 1.5 3 1022 cpm.
The peaks appear to be in pairs that have their third
wavenumbers (2ku 2 kw) in common.
c. Wavenumber–frequency slices
It is possible to include incorporate some useful frequency information in the wavenumber analysis without
resorting to full wavenumber–frequency bispectra. We
do so by bandpass filtering the high-frequency waves
into one-octave ranges: 6–12 cpd, 12–24 cpd, 24–48 cpd,
and 48–96 cpd, using sixth-order Butterworth filters
throughout. Because nearly all the energy transfers seen
in Figs. 14 and 15 are concentrated near the diagonal, we
compute a series of wavenumber bispectra, each using u9
and w9 from a single frequency range. As before, the
low-frequency waves include frequencies up to 3 cpd.
1542
JOURNAL OF PHYSICAL OCEANOGRAPHY
VOLUME 42
FIG. 17. Farfield: wavenumber bispectral estimates. As in Fig. 16, but for the Farfield. (left) Wavenumber bispectrum B (ku, kw): the
*
energy transfers are concentrated at very low wavenumbers, but the active region is elongated along the ku axis. (middle) Bicoherence
b(vu, vw): a wide band extending along the ku axis to high wavenumbers is above the bicoherence threshold. (Right) B*(ku, ku, 2ku 2 kw)
showing bicoherent values only: a contour plot is used to help locate the peak wavenumbers. Guide lines indicate constant sum wavenumbers (2ku 2 kw).
Figures 18 and 19 show the resulting wavenumber–
frequency slices. Bicoherence thresholding has already
been applied as in the rightmost panels of Figs. 16 and
17. As can be anticipated from the cumulative integrals
of energy transfer in Figs. 14 and 15, the strongest energy
transfers involve waves in the 24–48-cpd frequency
range. Some of the interactions can be partially localized in frequency: for example, in the Nearfield the
(0.9, 0.2, 21.1) 3 1022 cpm peak is seen strongly in the
24–48-cpd slice and less so in the 48–96-cpd slice, whereas
the (0.4, 0.2, 20.6) 3 1022 cpm interaction seems to be
found mostly in the 6–12- and 12–24-cpd slices. In the
Farfield, however, most of the peaks appear in much the
same locations regardless of frequency.
k
v2 2 f 2
5
k
N 2 2 v2
!1/2
.
(36)
The lowest frequency in the triad is v ’ f, so that the wave
vector k is nearly vertical. The sides k1, k2 correspond to
waves at higher frequencies (v1, v2) ’ (v1, v2 1 f).
As seen from the frequency bispectra (e.g., Fig. 14),
a typical value for v1 might be 24 cpd. Using values from
the Nearfield of f 5 0.74 cpd, N ’ 100 cpd, we find that
(v1, v2) ’ (32f, 33f ).
Because k ’ 0, the remaining sides have approximately equal horizontal wavenumber kH 5 k1 ’ k2. We
can estimate the relationship k1/k2 between their vertical wavenumbers using (36),
d. The interacting triads
Based on the bispectral analysis, we can form a general picture of the wave triads involved in the observed
energy transfers. The triangles in Fig. 20 depict example sum and difference triads in which the wave vectors,
k 5 (k, k), k1 5 (k1, k1), k2 5 (k2, k2), are coplanar.
If we assume linear dispersion (3), then the aspect
ratio k/k of each wave is determined by its frequency,
k1 /k2 ’ (k2 /k2 )/(k1 /k1 ) and
5
v22 2 f 2
N 2 2 v22
!,
!
v21 2 f 2
.
N 2 2 v21
(37)
(38)
For frequencies (32f, 33f ) we find k1/k2 ’ 1.04, meaning
that the vertical wavenumbers are nearly indistinguishable
SEPTEMBER 2012
SUN AND PINKEL
1543
FIG. 18. Nearfield: wavenumber–frequency bispectral slices. As in Fig. 16, but calculated for a series of frequency slices that include only
high-frequency waves in the bands 6–12 cpd, 12–24 cpd, 24–48 cpd, and 48–96 cpd. Some bispectral peaks appear to be localized in
frequency, but there appears to be significant leakage across slices.
from each other. In the sum reaction, the triangle is
nearly isosceles, and we recognize the elastic scattering
triad. The difference reaction is induced diffusion. In the
wavenumber bispectra, however, we find relatively large
ratios of k2/k1 . 2, which suggests that we are far from
the limiting cases of McComas and Bretherton (1977).
If resonant ID triads involve a pair of high-frequency
waves that differ only slightly in wavenumber, and ES
triads involve a pair of high-frequency waves that are
vertical near reflections of one another, then the interactions found here resemble neither case. Nor is it clear
that eikonal models of ID provide a useful explanation
for the observed «*, because the waves we find interacting are not small-scale packets propagating through
large-scale background shears but are even larger–scale
waves for which the shears do not represent a slowly
varying background in a ray-tracing sense.
One possibility for reconciling the observed frequencies v f with the large differences in aspect ratios is
that the high-frequency waves are actually waves that
have been Doppler shifted from low frequencies very
close to f, where a small change in frequency can lead to
a large change in aspect ratio. For example, one particular triad of frequencies that is familiar from studies
of parametric subharmonic instability, (v, v1, v2) 5
( f, M2 2 f, M2), has a ratio k1/k2 5 1.9. If the lowest
frequency in the triad is superinertial, then somewhat
higher ratios are possible: for example, (1.2f, M2 2 1.2f,
M2), gives a ratio k1/k2 5 2.5. However, we estimate that
a flow on the order of 3 m s21 would be required to
Doppler shift the ( f, M2 2 f, M2) triad so that the M2
wave is encountered at 24 cpd, where the inertial wave
has vertical wavenumber on the order of 1/ 100 cpm. Details of the calculation are found in the appendix. Given
FIG. 19. Farfield: wavenumber–frequency bispectral slices. As in Fig. 17, but calculated for a series of frequency slices that include only
high-frequency waves in the bands 6–12 cpd, 12–24 cpd, 24–48 cpd, and 48–96 cpd. In general, the bispectral peaks in the Farfield do not
appear to be well localized in frequency.
1544
JOURNAL OF PHYSICAL OCEANOGRAPHY
VOLUME 42
FIG. 20. Sum and difference triads. Example triads are diagrammed in vertical k and horizontal kH wavenumber space. In each case, k is the near-inertial (v ’ f ) member of the triad
and thus has a nearly vertical wave vector.
that barotropic flows in the Nearfield were on the order
of 30 cm s21, Doppler shifting of very low-frequency
waves to high encounter frequencies seems unlikely to
explain the triads we observe.
Because (36) is symmetric with respect to f and N,
another possibility is that the high-frequency waves
have intrinsic frequencies very close to N. For example,
k1/k2 5 2.1 for a triad with the same inertial wavenumber
as before and frequencies (f, N 2 1.3f, N 2 0.3f). In the
bottom 200 m of the Nearfield profiles, N ranges from
about 37 to 67 cpd, so that waves with frequency near N
would be detected as high-frequency waves even with
very little Doppler shifting downward in frequency.
However, this explanation would require that most of the
interactions involve waves close to the high-frequency
limit for internal waves. An additional difficulty is that N
increases with height, so that upward-propagating waves
that are near N at a given depth would quickly find
themselves far from N. In general, the validity of the
weakly resonant theory is unclear for waves propagating through significant changes in N within fractions
of a wavelength.
e. The energy transfer rate «* and turbulent
dissipation «
High-frequency internal waves are known eventually
to break, resulting in turbulent mixing (Alford and
Pinkel 2000a). How important are the observed energy
transfers to the high-frequency energy balance? To answer this question, we compare «* to independent estimates of the turbulent dissipation rate «.
Dissipation measurements were made at several sites
in HOME Nearfield using a combination of tethered
profilers and towed instruments. Directly over the ridge,
Klymak et al. (2006) reported averaged diffusivities on
top of the ridge of Kr ’ 1024 m2 s21 near the surface to
5 3 1023 m2 s21 near the bottom. These values correspond, through the Osborn (1980) relationship,
«
Kr 5 G *2 ,
hN i
(39)
to dissipations of « 5 8 3 1028 W kg21 and 2 3
1027 W kg21, using nominal values N 5 180 and
47 cpd, respectively. An analysis by Klymak et al. (2008)
of data collected aboard FLIP during HOME Nearfield
found similar values for « at the mooring site.
The apparent agreement in magnitude between «*
and « suggests that the interactions found here may be
of first-order importance to the energy balance between
internal waves and dissipation. On the other hand, much
of the deep turbulent mixing in the Nearfield has been
attributed to unrelated processes such as direct breaking
of the internal tide near the sloping bottom (Klymak et al.
2008), but agreement in the upper ocean is promising.
5. Summary
We report energy transfers from low-frequency, highshear internal waves to high-frequency waves in two
sites near Hawaii, using stress–shear triple correlations.
The internal wave field at the HOME Nearfield location
SEPTEMBER 2012
1545
SUN AND PINKEL
is characterized by elevated tidal amplitudes and shear
variance relative to open-ocean levels. Velocity spectra
are dominated by the semidiurnal tide and a set of
peaks at possible tidal harmonics, mostly associated with
upward-propagating wave energy. The HOME Farfield
wave field is somewhat less energetic and displays neither a marked vertical asymmetry nor prominent spectral peaks other than at near-inertial and semidiurnal
frequencies.
The energy transfer rate «* in the Nearfield is approximately 1.8 3 1027 W kg21, averaged from 100- to
700-m depths. In the Farfield, where the variances of
the correlated terms are 5 times smaller, averaged «* is
about a factor of 2 smaller, at 8.8 3 1028 W kg21.
Correlation coefficients r* are similar at 0.075 in the
Nearfield and 0.085 in the Farfield. Nearfield energy
transfers are largest in the lower part of the profiling
column where both shear and tidal amplitudes are large,
whereas «* is large in the Farfield near the surface.
Bispectral analysis demonstrates significantly bicoherent interactions between low-frequency waves and
a broad range of higher-frequency waves with similar
frequencies. Wavenumber bispectra find that the lowfrequency wave in each triad tends to have a wavelength
on the order of 100 m, as does one of the high-frequency
waves. However, the other high-frequency wave has
a much longer wavelength. The large mismatch between
the aspect ratios of the high-frequency waves is contrary
to classical definitions of induced diffusion or elastic
scattering interactions. Doppler shifting of the higherfrequency waves in each triad from either very low frequencies or very high frequencies does not appear to
explain the anomalous aspect ratios. The horizontal velocities required to shift waves with intrinsic frequencies
near f to observed frequencies of 24 cpd or higher are an
order of magnitude too large for tidal velocities, even in
the Nearfield.
Observed energy transfers in the Nearfield have similar magnitude to measurements of turbulent dissipation
near Kaena Ridge. The implication is that the interaction may be a significant step in the energy cascade
from low frequencies and scales near 100 m to breaking
waves and mixing. Although other pathways were seen
in HOME, including the direct breaking of near-seafloor
bores (Levine and Boyd 2006; Aucan et al. 2006; Klymak
et al. 2008) and the breaking of high-mode D2 lee waves
(Klymak et al. 2010), we feel that this pathway remains
active in the open ocean, as suggested by the finding of
a similar interaction in the Farfield, far from the ridge
slope.
Further study of the interaction is warranted. The
mismatch between the observed triads and the resonance condition is troubling, yet the energy transfer
rates appear robust. Why is this particular interaction
detectible relative to the classical wave–mean shear exchange? Which of the high-frequency waves is the principal recipient of the energy? Which is the most likely to
subsequently break?
Acknowledgments. The authors thank Eric Slater,
Mike Goldin, Lloyd Green, Mai Bui, Tyler Hughen,
Tony Aja, and Luc Rainville for support in the design of
instrumentation and collection of data at sea. Captain
Tom Golfinos and the crew of the research platform
FLIP provided exemplary support. Discussions with Bill
Young, Jen MacKinnon, Jody Klymak, Kurt Polzin, and
Wen-Yih Sun were helpful. This work was supported
by the National Science Foundation and the Office of
Naval Research.
APPENDIX
Doppler Shifting and Resonant Triads
The frequency at which we encounter (measure) a
given wave that is being advected in a flow will differ from
its intrinsic frequency, which is the frequency we would
measure if there were no motion (or, equivalently, if
we were moving with the flow). The frequency v that we
measure in an Eulerian frame can be related v to the
intrinsic frequency v
^ and the flow velocity u through
v5v
^ 1 (k u).
(A1)
The Doppler shift (k u) is due to the advection of
spatial phase variations k past the fixed observer by the
flow u.
The resonance conditions (1) require that a resonant
triad maintain fixed mutual phase in both space and
time, so Doppler shifting by an external velocity field is
not expected to affect the existence of resonant energy
transfers, as all three waves will be advected in unison.
However, waves can be shifted out of a given frequency
band in which we had expected to find them. For
example, the wavenumber–frequency shear spectra of
shears (Figs. 5, 6) show the effects of Doppler shifting
as K-shaped patterns that originate at low frequency
and widen with increasing wavenumber. As discussed
by Pinkel (2008), this is spreading of variance is partially
due (tidal) vertical heaving of high-vertical-wavenumber
shears. As a result, some of the low-frequency shear
variance is encountered at high frequencies and falls
outside our filter band for low-frequency (›U/›z). The
weighting of the spectrum toward low frequencies suggests that more variance is Doppler shifted out of the
low-frequency band than is been shifted in from other
frequencies.
1546
JOURNAL OF PHYSICAL OCEANOGRAPHY
Doppler shifts by vertical heaving cannot explain
the observed frequencies of the triads. The near-inertial
(observed) waves in each triad tend to have the largest vertical wavenumbers and would be easily Doppler
shifted out of the low-frequency band by vertical motions. Consider instead Doppler shifting of the triad
discussed above by a horizontal flow u 5 u such as the
barotropic tide. The near-inertial member of the triad,
with nearly zero horizontal wavenumber k ’ 0, is relatively unaffected by the horizontal flow. The other two
members of the triad, sharing the same horizontal
wavenumber kh, are Doppler shifted by equal amounts.
How strong of a flow is needed to Doppler-shift the
M2 wave in the (f, M2 2 f, M2) triad to an observed frequency of 24 cpd? From Fig. 16, we can estimate a vertical
wavenumber of k 5 1/100 cpm for the inertial wave.
Using k1/k2 5 1.9 and (36), we find a shared horizontal
wavenumber for the M2 2 f, M2 waves of kH 5 7.7 3
1025 cpm. Then, using (A1),
24 cpd 5 M2 1 kH u,
2:78 3 1024 s21 ’ 2:2 3 1025 s21 1 (7:7 3 1025 m21 )u,
and
u ’ 3:3 m s21 .
REFERENCES
Alford, M., and R. Pinkel, 2000a: Observations of overturning in
the thermocline: The context of ocean mixing. J. Phys. Oceanogr., 30, 805–832.
——, and ——, 2000b: Patterns of turbulent and double-diffusive
phenomena: Observations from a rapid-profiling microconductivity probe. J. Phys. Oceanogr., 30, 833–854.
——, and M. C. Gregg, 2001: Near-inertial mixing: Modulation of
shear, strain and microstructure at low latitude. J. Geophys.
Res., 106 (C8), 16 947–16 968.
Aucan, J., M. A. Merrifield, D. S. Luther, and P. Flament, 2006:
Tidal mixing events on the deep flanks of Kaena Ridge,
Hawaii. J. Phys. Oceanogr., 36, 1202–1219.
Benney, D., and P. Saffman, 1966: Nonlinear interactions of random waves in a dispersive medium. Proc. Roy. Soc. London,
289A, 301–320.
Broutman, D., and W. Young, 1986: On the interaction of smallscale oceanic internal waves with near-inertial waves. J. Fluid
Mech., 166, 341–358.
Carter, G., and M. Gregg, 2006: Persistent near-diurnal internal
waves observed above a site of M2 barotropic-to-baroclinic
conversion. J. Phys. Oceanogr., 36, 1136–1147.
Duda, T., and D. Jacobs, 1998: Stress/shear correlation: Internal
wave/wave interaction and energy flux in the upper ocean.
Geophys. Res. Lett., 25, 1919–1922.
Elgar, S., and R. Guza, 1985: Shoaling gravity waves: Comparisons
between field observations, linear theory, and a nonlinear
model. J. Fluid Mech., 158, 47–70.
VOLUME 42
——, and ——, 1988: Statistics of bicoherence. IEEE Trans. Acoust.
Speech Signal Process., 36, 1667–1668.
——, T. Herbers, V. Chandran, and R. Guza, 1995: Higher-order
spectral-analysis of nonlinear ocean surface gravity-waves.
J. Geophys. Res., 100 (C3), 4977–4983.
Frajka-Williams, E. E., E. L. Kunze, and J. A. MacKinnon, 2006:
Bispectra of internal tides and parametric subharmonic instability. M.S. thesis, School of Oceanography, University of
Washington, 26 pp.
Furue, R., 2003: Energy transfer within the small-scale oceanic
internal wave spectrum. J. Phys. Oceanogr., 33, 267–282.
Furuichi, N., T. Hibiya, and Y. Niwa, 2005: Bispectral analysis of
energy transfer within the two-dimensional oceanic internal
wave field. J. Phys. Oceanogr., 35, 2104–2109.
Gargett, A., and G. Holloway, 1984: Dissipation and diffusion by
internal wave breaking. J. Mar. Res., 42, 15–27.
Gill, A. E., 1982: Atmosphere–Ocean Dynamics. Academic Press,
662 pp.
Gregg, M. C., 1989: Scaling turbulent dissipation in the thermocline. J. Geophys. Res., 94 (C7), 9686–9698.
Hasselmann, K., 1962: On the non-linear energy transfer in
a gravity-wave spectrum. 1. General theory. J. Fluid Mech., 12,
481–500.
Henyey, F., and N. Pomphrey, 1983: Eikonal description of internal
wave interactions—A non-diffusive picture of induced diffusion. Dyn. Atmos. Oceans, 7, 189–219.
——, J. Wright, and S. Flatté, 1986: Energy and action flow through
the internal wave field: An eikonal approach. J. Geophys. Res.,
91 (C7), 8487–8495.
Hinich, M., and C. Clay, 1968: The application of the discrete
fourier transform in the estimation of power spectra, coherence, and bispectra of geophysical data. Rev. Geophys.
Space Phys., 6, 289–346.
——, and M. Wolinsky, 2005: Normalizing bispectra. J. Stat. Plann.
Inference, 130 (1–2), 405–411.
Jacobs, D. C., and C. S. Cox, 1987: Internal wave stress-shear
correlations: A choice of reference frames. Geophys. Res.
Lett., 14 (1), 25–28.
Kim, Y., and E. Powers, 1979: Digital bispectral analysis and its
applications to non-linear wave interactions. IEEE Trans.
Plasma Sci., 7, 120–131.
Klymak, J. M., and Coauthors, 2006: An estimate of tidal energy
lost to turbulence at the Hawaiian Ridge. J. Phys. Oceanogr.,
36, 1148–1164.
——, R. Pinkel, and L. Rainville, 2008: Direct breaking of the internal tide near topography: Kaena Ridge, Hawaii. J. Phys.
Oceanogr., 38, 380–399.
——, S. M. Legg, and R. Pinkel, 2010: High-mode stationary waves
in stratified flow over large obstacles. J. Fluid Mech., 644,
321–366.
Levine, M. D., and T. J. Boyd, 2006: Tidally forced internal waves
and overturns observed on a slope: Results from HOME.
J. Phys. Oceanogr., 36, 1184–1201.
McComas, C. H., and F. Bretherton, 1977: Resonant interaction
of oceanic internal waves. J. Geophys. Res., 82 (9), 1397–
1412.
——, and M. Briscoe, 1980: Bispectra of internal waves. J. Fluid
Mech., 97, 205–213.
——, and P. Müller, 1981: The dynamic balance of internal waves.
J. Phys. Oceanogr., 11, 970–986.
Müller, P., G. Holloway, and F. Henyey, 1986: Nonlinear interactions among internal gravity waves. Rev. Geophys., 24,
493–536.
SEPTEMBER 2012
SUN AND PINKEL
Munk, W., 1981: Internal waves and small-scale processes. Evolution of Physical Oceanography, B. A. Warren and C. Wunsch,
Eds., MIT Press, 264–290.
Osborn, T. R., 1980: Estimates of the local rate of vertical diffusion from dissipation measurements. J. Phys. Oceanogr., 10,
83–89.
Phillips, O. M., 1960: On the dynamics of unsteady gravity waves of
finite amplitude. J. Fluid Mech., 9, 193–217.
Pinkel, R., 2008: Advection, phase distortion, and the frequency
spectrum of finescale fields in the sea. J. Phys. Oceanogr., 38,
291–313.
——, and J. Smith, 1992: Repeat-sequence coding for improved
precision of Doppler sonar and sodar. J. Atmos. Oceanic
Technol., 9, 149–163.
1547
Polzin, K., J. Toole, and R. Schmitt, 1995: Finescale parameterizations of turbulent dissipation. J. Phys. Oceanogr., 25, 306–328.
Rainville, L., and R. Pinkel, 2006a: Baroclinic energy flux at the
Hawaiian Ridge: Observations from the R/P FLIP. J. Phys.
Oceanogr., 36, 1104–1122.
——, and ——, 2006b: Propagation of low-mode internal waves
through the ocean. J. Phys. Oceanogr., 36, 1220–1236.
Ruddick, B., and T. Joyce, 1979: Observations of interaction between the internal wave-field and low-frequency flows in the
North Atlantic. J. Phys. Oceanogr., 9, 498–517.
Rudnick, D. L., and Coauthors, 2003: From tides to mixing along
the Hawaiian Ridge. Science, 301, 355–357.
Tennekes, H., and J. L. Lumley, 1972: A First Course in Turbulence.
MIT Press, 300 pp.
Download